blob: c2b41a85013eec87a412badc4097f364d547726d [file] [log] [blame]
Junio C Hamanof859c842007-05-11 22:11:07 -07001#include "cache.h"
2
Junio C Hamano72f31962012-09-15 22:38:28 -07003static int threaded_check_leading_path(struct cache_def *cache, const char *name, int len);
4static int threaded_has_dirs_only_path(struct cache_def *cache, const char *name, int len, int prefix_len);
5
Kjetil Barvik148bc062009-02-09 21:54:05 +01006/*
7 * Returns the length (on a path component basis) of the longest
8 * common prefix match of 'name_a' and 'name_b'.
9 */
10static int longest_path_match(const char *name_a, int len_a,
11 const char *name_b, int len_b,
12 int *previous_slash)
13{
14 int max_len, match_len = 0, match_len_prev = 0, i = 0;
15
16 max_len = len_a < len_b ? len_a : len_b;
17 while (i < max_len && name_a[i] == name_b[i]) {
18 if (name_a[i] == '/') {
19 match_len_prev = match_len;
20 match_len = i;
21 }
22 i++;
23 }
24 /*
25 * Is 'name_b' a substring of 'name_a', the other way around,
26 * or is 'name_a' and 'name_b' the exact same string?
27 */
28 if (i >= max_len && ((len_a > len_b && name_a[len_b] == '/') ||
29 (len_a < len_b && name_b[len_a] == '/') ||
30 (len_a == len_b))) {
31 match_len_prev = match_len;
32 match_len = i;
33 }
34 *previous_slash = match_len_prev;
35 return match_len;
36}
37
Linus Torvaldsb9fd2842009-07-09 13:35:31 -070038static struct cache_def default_cache;
Junio C Hamanof859c842007-05-11 22:11:07 -070039
Linus Torvalds867f72b2009-07-09 13:23:59 -070040static inline void reset_lstat_cache(struct cache_def *cache)
Linus Torvaldsc40641b2008-05-09 09:21:07 -070041{
Linus Torvalds867f72b2009-07-09 13:23:59 -070042 cache->path[0] = '\0';
43 cache->len = 0;
44 cache->flags = 0;
Kjetil Barvik60b458b2009-02-09 21:54:04 +010045 /*
46 * The track_flags and prefix_len_stat_func members is only
47 * set by the safeguard rule inside lstat_cache()
48 */
Kjetil Barvik92604b42009-01-18 16:14:50 +010049}
50
51#define FL_DIR (1 << 0)
Kjetil Barvik09c93062009-01-18 16:14:51 +010052#define FL_NOENT (1 << 1)
53#define FL_SYMLINK (1 << 2)
54#define FL_LSTATERR (1 << 3)
55#define FL_ERR (1 << 4)
Kjetil Barvikbad4a542009-01-18 16:14:52 +010056#define FL_FULLPATH (1 << 5)
Kjetil Barvik92604b42009-01-18 16:14:50 +010057
58/*
59 * Check if name 'name' of length 'len' has a symlink leading
Kjetil Barvik09c93062009-01-18 16:14:51 +010060 * component, or if the directory exists and is real, or not.
Kjetil Barvik92604b42009-01-18 16:14:50 +010061 *
62 * To speed up the check, some information is allowed to be cached.
Kjetil Barvikbad4a542009-01-18 16:14:52 +010063 * This can be indicated by the 'track_flags' argument, which also can
64 * be used to indicate that we should check the full path.
65 *
66 * The 'prefix_len_stat_func' parameter can be used to set the length
67 * of the prefix, where the cache should use the stat() function
68 * instead of the lstat() function to test each path component.
Kjetil Barvik92604b42009-01-18 16:14:50 +010069 */
Clemens Buchacher4856ff22010-10-09 15:52:59 +020070static int lstat_cache_matchlen(struct cache_def *cache,
71 const char *name, int len,
72 int *ret_flags, int track_flags,
73 int prefix_len_stat_func)
Kjetil Barvik92604b42009-01-18 16:14:50 +010074{
Kjetil Barvikaeabab52009-01-18 16:14:53 +010075 int match_len, last_slash, last_slash_dir, previous_slash;
Clemens Buchacher4856ff22010-10-09 15:52:59 +020076 int save_flags, max_len, ret;
Linus Torvaldsc40641b2008-05-09 09:21:07 -070077 struct stat st;
Junio C Hamanof859c842007-05-11 22:11:07 -070078
Linus Torvalds867f72b2009-07-09 13:23:59 -070079 if (cache->track_flags != track_flags ||
80 cache->prefix_len_stat_func != prefix_len_stat_func) {
Kjetil Barvik09c93062009-01-18 16:14:51 +010081 /*
Kjetil Barvik60b458b2009-02-09 21:54:04 +010082 * As a safeguard rule we clear the cache if the
83 * values of track_flags and/or prefix_len_stat_func
84 * does not match with the last supplied values.
Kjetil Barvik09c93062009-01-18 16:14:51 +010085 */
Linus Torvalds867f72b2009-07-09 13:23:59 -070086 reset_lstat_cache(cache);
87 cache->track_flags = track_flags;
88 cache->prefix_len_stat_func = prefix_len_stat_func;
Kjetil Barvik09c93062009-01-18 16:14:51 +010089 match_len = last_slash = 0;
90 } else {
91 /*
92 * Check to see if we have a match from the cache for
93 * the 2 "excluding" path types.
94 */
Kjetil Barvikaeabab52009-01-18 16:14:53 +010095 match_len = last_slash =
Linus Torvalds867f72b2009-07-09 13:23:59 -070096 longest_path_match(name, len, cache->path, cache->len,
Kjetil Barvik148bc062009-02-09 21:54:05 +010097 &previous_slash);
Clemens Buchacher4856ff22010-10-09 15:52:59 +020098 *ret_flags = cache->flags & track_flags & (FL_NOENT|FL_SYMLINK);
Kjetil Barvik77716752009-06-14 15:08:28 +020099
100 if (!(track_flags & FL_FULLPATH) && match_len == len)
101 match_len = last_slash = previous_slash;
102
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200103 if (*ret_flags && match_len == cache->len)
104 return match_len;
Kjetil Barvik09c93062009-01-18 16:14:51 +0100105 /*
106 * If we now have match_len > 0, we would know that
107 * the matched part will always be a directory.
108 *
109 * Also, if we are tracking directories and 'name' is
110 * a substring of the cache on a path component basis,
111 * we can return immediately.
112 */
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200113 *ret_flags = track_flags & FL_DIR;
114 if (*ret_flags && len == match_len)
115 return match_len;
Kjetil Barvik09c93062009-01-18 16:14:51 +0100116 }
Linus Torvaldsc40641b2008-05-09 09:21:07 -0700117
118 /*
Kjetil Barvik92604b42009-01-18 16:14:50 +0100119 * Okay, no match from the cache so far, so now we have to
120 * check the rest of the path components.
Linus Torvaldsc40641b2008-05-09 09:21:07 -0700121 */
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200122 *ret_flags = FL_DIR;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100123 last_slash_dir = last_slash;
124 max_len = len < PATH_MAX ? len : PATH_MAX;
125 while (match_len < max_len) {
126 do {
Linus Torvalds867f72b2009-07-09 13:23:59 -0700127 cache->path[match_len] = name[match_len];
Kjetil Barvik92604b42009-01-18 16:14:50 +0100128 match_len++;
129 } while (match_len < max_len && name[match_len] != '/');
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100130 if (match_len >= max_len && !(track_flags & FL_FULLPATH))
Kjetil Barvik92604b42009-01-18 16:14:50 +0100131 break;
132 last_slash = match_len;
Linus Torvalds867f72b2009-07-09 13:23:59 -0700133 cache->path[last_slash] = '\0';
Linus Torvaldsc40641b2008-05-09 09:21:07 -0700134
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100135 if (last_slash <= prefix_len_stat_func)
Linus Torvalds867f72b2009-07-09 13:23:59 -0700136 ret = stat(cache->path, &st);
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100137 else
Linus Torvalds867f72b2009-07-09 13:23:59 -0700138 ret = lstat(cache->path, &st);
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100139
140 if (ret) {
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200141 *ret_flags = FL_LSTATERR;
Kjetil Barvik09c93062009-01-18 16:14:51 +0100142 if (errno == ENOENT)
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200143 *ret_flags |= FL_NOENT;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100144 } else if (S_ISDIR(st.st_mode)) {
145 last_slash_dir = last_slash;
Linus Torvaldsc40641b2008-05-09 09:21:07 -0700146 continue;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100147 } else if (S_ISLNK(st.st_mode)) {
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200148 *ret_flags = FL_SYMLINK;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100149 } else {
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200150 *ret_flags = FL_ERR;
Junio C Hamanof859c842007-05-11 22:11:07 -0700151 }
Linus Torvaldsc40641b2008-05-09 09:21:07 -0700152 break;
Junio C Hamanof859c842007-05-11 22:11:07 -0700153 }
Kjetil Barvik92604b42009-01-18 16:14:50 +0100154
155 /*
Kjetil Barvik09c93062009-01-18 16:14:51 +0100156 * At the end update the cache. Note that max 3 different
157 * path types, FL_NOENT, FL_SYMLINK and FL_DIR, can be cached
158 * for the moment!
Kjetil Barvik92604b42009-01-18 16:14:50 +0100159 */
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200160 save_flags = *ret_flags & track_flags & (FL_NOENT|FL_SYMLINK);
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100161 if (save_flags && last_slash > 0 && last_slash <= PATH_MAX) {
Linus Torvalds867f72b2009-07-09 13:23:59 -0700162 cache->path[last_slash] = '\0';
163 cache->len = last_slash;
164 cache->flags = save_flags;
Kjetil Barvik60b458b2009-02-09 21:54:04 +0100165 } else if ((track_flags & FL_DIR) &&
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100166 last_slash_dir > 0 && last_slash_dir <= PATH_MAX) {
Kjetil Barvik92604b42009-01-18 16:14:50 +0100167 /*
168 * We have a separate test for the directory case,
Kjetil Barvik09c93062009-01-18 16:14:51 +0100169 * since it could be that we have found a symlink or a
170 * non-existing directory and the track_flags says
171 * that we cannot cache this fact, so the cache would
172 * then have been left empty in this case.
Kjetil Barvik92604b42009-01-18 16:14:50 +0100173 *
174 * But if we are allowed to track real directories, we
175 * can still cache the path components before the last
Kjetil Barvik09c93062009-01-18 16:14:51 +0100176 * one (the found symlink or non-existing component).
Kjetil Barvik92604b42009-01-18 16:14:50 +0100177 */
Linus Torvalds867f72b2009-07-09 13:23:59 -0700178 cache->path[last_slash_dir] = '\0';
179 cache->len = last_slash_dir;
180 cache->flags = FL_DIR;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100181 } else {
Linus Torvalds867f72b2009-07-09 13:23:59 -0700182 reset_lstat_cache(cache);
Kjetil Barvik92604b42009-01-18 16:14:50 +0100183 }
Clemens Buchacher4856ff22010-10-09 15:52:59 +0200184 return match_len;
185}
186
187static int lstat_cache(struct cache_def *cache, const char *name, int len,
188 int track_flags, int prefix_len_stat_func)
189{
190 int flags;
191 (void)lstat_cache_matchlen(cache, name, len, &flags, track_flags,
192 prefix_len_stat_func);
193 return flags;
Kjetil Barvik92604b42009-01-18 16:14:50 +0100194}
195
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100196#define USE_ONLY_LSTAT 0
197
Kjetil Barvik92604b42009-01-18 16:14:50 +0100198/*
199 * Return non-zero if path 'name' has a leading symlink component
200 */
Linus Torvaldsb9fd2842009-07-09 13:35:31 -0700201int threaded_has_symlink_leading_path(struct cache_def *cache, const char *name, int len)
202{
203 return lstat_cache(cache, name, len, FL_SYMLINK|FL_DIR, USE_ONLY_LSTAT) & FL_SYMLINK;
204}
205
206/*
207 * Return non-zero if path 'name' has a leading symlink component
208 */
Kjetil Barvik57199892009-02-09 21:54:06 +0100209int has_symlink_leading_path(const char *name, int len)
Kjetil Barvik92604b42009-01-18 16:14:50 +0100210{
Linus Torvaldsb9fd2842009-07-09 13:35:31 -0700211 return threaded_has_symlink_leading_path(&default_cache, name, len);
Junio C Hamanof859c842007-05-11 22:11:07 -0700212}
Kjetil Barvik09c93062009-01-18 16:14:51 +0100213
214/*
Clemens Buchacherf66caaf2010-10-09 15:53:00 +0200215 * Return zero if path 'name' has a leading symlink component or
Kjetil Barvik09c93062009-01-18 16:14:51 +0100216 * if some leading path component does not exists.
Clemens Buchacherf66caaf2010-10-09 15:53:00 +0200217 *
218 * Return -1 if leading path exists and is a directory.
219 *
220 * Return path length if leading path exists and is neither a
221 * directory nor a symlink.
Kjetil Barvik09c93062009-01-18 16:14:51 +0100222 */
Clemens Buchacherf66caaf2010-10-09 15:53:00 +0200223int check_leading_path(const char *name, int len)
Kjetil Barvik09c93062009-01-18 16:14:51 +0100224{
Jared Hance15438d52012-03-02 21:31:15 -0500225 return threaded_check_leading_path(&default_cache, name, len);
226}
227
228/*
229 * Return zero if path 'name' has a leading symlink component or
230 * if some leading path component does not exists.
231 *
232 * Return -1 if leading path exists and is a directory.
233 *
234 * Return path length if leading path exists and is neither a
235 * directory nor a symlink.
236 */
Junio C Hamano72f31962012-09-15 22:38:28 -0700237static int threaded_check_leading_path(struct cache_def *cache, const char *name, int len)
Jared Hance15438d52012-03-02 21:31:15 -0500238{
Clemens Buchacherf66caaf2010-10-09 15:53:00 +0200239 int flags;
240 int match_len = lstat_cache_matchlen(cache, name, len, &flags,
241 FL_SYMLINK|FL_NOENT|FL_DIR, USE_ONLY_LSTAT);
Clemens Buchacher1d718a52011-02-20 13:13:43 +0100242 if (flags & FL_NOENT)
Clemens Buchacherf66caaf2010-10-09 15:53:00 +0200243 return 0;
244 else if (flags & FL_DIR)
245 return -1;
246 else
247 return match_len;
Kjetil Barvik09c93062009-01-18 16:14:51 +0100248}
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100249
250/*
251 * Return non-zero if all path components of 'name' exists as a
252 * directory. If prefix_len > 0, we will test with the stat()
253 * function instead of the lstat() function for a prefix length of
254 * 'prefix_len', thus we then allow for symlinks in the prefix part as
255 * long as those points to real existing directories.
256 */
Kjetil Barvik57199892009-02-09 21:54:06 +0100257int has_dirs_only_path(const char *name, int len, int prefix_len)
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100258{
Jared Hance15438d52012-03-02 21:31:15 -0500259 return threaded_has_dirs_only_path(&default_cache, name, len, prefix_len);
260}
261
262/*
263 * Return non-zero if all path components of 'name' exists as a
264 * directory. If prefix_len > 0, we will test with the stat()
265 * function instead of the lstat() function for a prefix length of
266 * 'prefix_len', thus we then allow for symlinks in the prefix part as
267 * long as those points to real existing directories.
268 */
Junio C Hamano72f31962012-09-15 22:38:28 -0700269static int threaded_has_dirs_only_path(struct cache_def *cache, const char *name, int len, int prefix_len)
Jared Hance15438d52012-03-02 21:31:15 -0500270{
Linus Torvalds867f72b2009-07-09 13:23:59 -0700271 return lstat_cache(cache, name, len,
Kjetil Barvikbad4a542009-01-18 16:14:52 +0100272 FL_DIR|FL_FULLPATH, prefix_len) &
273 FL_DIR;
274}
Kjetil Barvik78478922009-02-09 21:54:07 +0100275
276static struct removal_def {
277 char path[PATH_MAX];
278 int len;
279} removal;
280
281static void do_remove_scheduled_dirs(int new_len)
282{
283 while (removal.len > new_len) {
284 removal.path[removal.len] = '\0';
285 if (rmdir(removal.path))
286 break;
287 do {
288 removal.len--;
289 } while (removal.len > new_len &&
290 removal.path[removal.len] != '/');
291 }
292 removal.len = new_len;
Kjetil Barvik78478922009-02-09 21:54:07 +0100293}
294
295void schedule_dir_for_removal(const char *name, int len)
296{
297 int match_len, last_slash, i, previous_slash;
298
299 match_len = last_slash = i =
300 longest_path_match(name, len, removal.path, removal.len,
301 &previous_slash);
302 /* Find last slash inside 'name' */
303 while (i < len) {
304 if (name[i] == '/')
305 last_slash = i;
306 i++;
307 }
308
309 /*
310 * If we are about to go down the directory tree, we check if
311 * we must first go upwards the tree, such that we then can
312 * remove possible empty directories as we go upwards.
313 */
314 if (match_len < last_slash && match_len < removal.len)
315 do_remove_scheduled_dirs(match_len);
316 /*
317 * If we go deeper down the directory tree, we only need to
318 * save the new path components as we go down.
319 */
320 if (match_len < last_slash) {
321 memcpy(&removal.path[match_len], &name[match_len],
322 last_slash - match_len);
323 removal.len = last_slash;
324 }
Kjetil Barvik78478922009-02-09 21:54:07 +0100325}
326
327void remove_scheduled_dirs(void)
328{
329 do_remove_scheduled_dirs(0);
Kjetil Barvik78478922009-02-09 21:54:07 +0100330}